Nanopore bulk RNA-Seq and scRNA-Seq data benchmarks (top 4000 highly variable transcripts)
Preparing the data
Load the required packages:
library(tidyverse)
library(glue)
library(scales)
library(Matrix)
library(scater)
library(scran)
library(pheatmap)
library(viridis)Global parameters:
csv_dir <- "nanopore_bulk_sc_benchmarks_hvts_data"
dir.create(csv_dir, recursive = TRUE)
bulk_dir <- file.path("..", "nanopore_bulk_analysis", "report_data")
sc_dir <- file.path("..", "nanopore_sc_analysis", "report_data")
souporcell_file <- file.path("..", "illumina_sc_analysis", "data",
"souporcell_clusters.tsv")
bulk_sample_ids <- c("LIB5432309_SAM24385452", "LIB5432310_SAM24385453",
"LIB5432311_SAM24385454", "LIB5432312_SAM24385455",
"LIB5432313_SAM24385456", "LIB5432314_SAM24385457",
"LIB5432315_SAM24385458", "LIB5432316_SAM24385459")
bulk_sample_names <- c("OVMANA", "OVKATE", "OVTOKO", "SK-OV-3", "COV362",
"COV504", "IGROV-1")
bulk_sample_order <- c("SK-OV-3", "IGROV-1", "COV504", "OVMANA", "OVKATE",
"OVTOKO", "COV362")
cell_min_txs <- 500
top_n_hvts <- 4000
min_cor_spearman <- 0.3
max_cor_spearman <- 0.9
min_mean_rel_diff <- 0.5
max_mean_rel_diff <- 1.1
heatmap_palette <- cividis
heatmap_palette_direction <- 1
scatterplot_df_list <- list()Helper functions:
fill_missing_matrix <- function(x, all_rownames) {
missing_rownames <- setdiff(all_rownames, rownames(x))
missing_matrix <- matrix(
0, nrow = length(missing_rownames), ncol = ncol(x)
)
rownames(missing_matrix) <- missing_rownames
full_matrix <- rbind(x, missing_matrix)
full_matrix <- full_matrix[all_rownames,]
return(full_matrix)
}
get_hvts <- function(count_matrix) {
sce <- SingleCellExperiment(assays = list(counts = count_matrix))
sce <- sce[, colSums(count_matrix > 0) >= cell_min_txs]
sce <- computeLibraryFactors(sce)
sce <- logNormCounts(sce)
dec <- modelGeneVar(sce)
top_hvts <- getTopHVGs(dec, n = top_n_hvts)
return(top_hvts)
}
get_mean_rel_diff <- function(x, y) {
selector <- (x != 0) & (y != 0)
x <- x[selector]
y <- y[selector]
mean(abs(x - y) / ((x + y) / 2))
}
get_mean_rel_diff_matrix <- function(x, y) {
apply(y, 2, function(y_values) {
apply(x, 2, function(x_values) {
selector <- (y_values != 0) & (x_values != 0)
y_values <- y_values[selector]
x_values <- x_values[selector]
mean(abs(x_values - y_values) / ((x_values + y_values) / 2))
})
})
}
prepare_scatterplot_data <- function(tpm_matrix_bulk,
tpm_matrix_pseudobulk) {
tpm_bulk_df <- as.data.frame(tpm_matrix_bulk[, 1:3])
bulk_levels <- colnames(tpm_bulk_df)
tpm_bulk_df$transcript_id <- rownames(tpm_bulk_df)
rownames(tpm_bulk_df) <- NULL
tpm_bulk_df <- tpm_bulk_df %>%
gather(-transcript_id, key = "sample_bulk", value = "tpm_bulk")
tpm_bulk_df$sample_bulk <- factor(
tpm_bulk_df$sample_bulk, levels = bulk_levels
)
tpm_pseudobulk_df <- as.data.frame(tpm_matrix_pseudobulk)
pseudobulk_levels <- colnames(tpm_pseudobulk_df)
tpm_pseudobulk_df$transcript_id <- rownames(tpm_pseudobulk_df)
rownames(tpm_pseudobulk_df) <- NULL
tpm_pseudobulk_df <- tpm_pseudobulk_df %>%
gather(-transcript_id, key = "sample_pseudobulk",
value = "tpm_pseudobulk")
tpm_pseudobulk_df$sample_pseudobulk <- factor(
tpm_pseudobulk_df$sample_pseudobulk, levels = pseudobulk_levels
)
tpm_df <- left_join(tpm_bulk_df, tpm_pseudobulk_df)
tpm_df$status <- ifelse(
as.numeric(tpm_df$sample_bulk) ==
as.numeric(tpm_df$sample_pseudobulk),
"Correct vs Correct", "Correct vs Decoy"
)
tpm_df$status <- factor(
tpm_df$status,
levels = c("Correct vs Correct", "Correct vs Decoy")
)
return(tpm_df)
}
create_scatterplot_paired <- function(plot_df) {
plot_df$tpm_bulk[plot_df$tpm_bulk < 0.1] <- 0.1
plot_df$tpm_pseudobulk[plot_df$tpm_pseudobulk < 0.1] <- 0.1
scatter_plot <- ggplot(plot_df, mapping = aes(x = tpm_bulk,
y = tpm_pseudobulk)) +
geom_point(alpha = 0.35, size = 0.05) +
geom_abline(intercept = 0, slope = 1,
lty = 3, color = "red") +
facet_grid(rows = vars(sample_pseudobulk),
cols = vars(sample_bulk)) +
scale_x_log10(labels = trans_format("log10", math_format(10^.x))) +
scale_y_log10(labels = trans_format("log10", math_format(10^.x))) +
coord_cartesian(xlim = c(0.1, 10^5),
ylim = c(0.1, 10^5),
clip = "off") +
annotation_logticks(outside = TRUE,
size = 0.25,
short = unit(0.05, "cm"),
mid = unit(0.1, "cm"),
long = unit(0.15, "cm")) +
theme_bw() +
theme(aspect.ratio = 1,
axis.title.x = element_blank(),
axis.title.y = element_blank())
return(scatter_plot)
}
create_scatterplot_combined <- function(plot_df) {
plot_df$tpm_bulk[plot_df$tpm_bulk < 0.1] <- 0.1
plot_df$tpm_pseudobulk[plot_df$tpm_pseudobulk < 0.1] <- 0.1
scatter_plot <- ggplot(plot_df, mapping = aes(x = tpm_bulk,
y = tpm_pseudobulk)) +
geom_point(alpha = 0.35, size = 0.05) +
geom_abline(intercept = 0, slope = 1,
lty = 3, color = "red") +
facet_grid(cols = vars(status)) +
scale_x_log10(labels = trans_format("log10", math_format(10^.x))) +
scale_y_log10(labels = trans_format("log10", math_format(10^.x))) +
coord_cartesian(xlim = c(0.1, 10^5),
ylim = c(0.1, 10^5),
clip = "off") +
annotation_logticks(outside = TRUE,
size = 0.25,
short = unit(0.05, "cm"),
mid = unit(0.1, "cm"),
long = unit(0.15, "cm")) +
theme_bw() +
theme(aspect.ratio = 1,
axis.title.x = element_blank(),
axis.title.y = element_blank())
return(scatter_plot)
}Prepare mitochondrial transcript blacklist:
se_transcript_mt <- readRDS(file.path(
sc_dir, "isosceles_se_transcript.rds"
))
se_transcript_mt <- se_transcript_mt[
grepl("^MT:", rowData(se_transcript_mt)$position),
]
mt_transcripts_isosceles <- rownames(se_transcript_mt)
mt_transcripts_ensembl <- rowData(se_transcript_mt)$compatible_tx
mt_transcripts_isosceles <- c(
mt_transcripts_isosceles,
c("ISOT-0000-0000-0000-2622:s10197250:e10199150:FA:FL",
"ISOT-0000-0000-0000-caeb:s10507850:e10509200:FA:FL",
"ISOT-0000-0000-0000-2809:s96617150:e96618250:FA:FL")
)
mt_transcripts_ensembl <- c(
mt_transcripts_ensembl,
c("ENST00000445125", "ENST00000536684", "ENST00000600213")
)Prepare Souporcell cluster data:
souporcell_df <- read.delim(souporcell_file)
souporcell_barcode <- gsub("-1$", "", souporcell_df$barcode)
souporcell_cluster <- ifelse(souporcell_df$status == "singlet",
souporcell_df$assignment,
souporcell_df$status)
souporcell_cluster <- setNames(souporcell_cluster, souporcell_barcode)Isosceles bulk-pseudobulk correlation plots
Prepare Isosceles scRNA-Seq data:
se <- readRDS(file.path(
sc_dir, "isosceles_se_transcript.rds"
))
colnames(se) <- sapply(strsplit(colnames(se), "\\."), "[", 2)
sc_counts <- assay(se, "counts")
sc_counts <- sc_counts[
!(rownames(sc_counts) %in% mt_transcripts_isosceles),
]
cell_labels <- souporcell_cluster[colnames(sc_counts)]
cell_selector <- cell_labels %in% c("0", "1", "2")
sc_counts <- sc_counts[, cell_selector]Prepare Isosceles pseudobulk RNA-Seq data:
se_pseudobulk <- readRDS(file.path(
sc_dir, "isosceles_se_pseudobulk_transcript.rds"
))
se_pseudobulk <- se_pseudobulk[, 1:3]
colnames(se_pseudobulk) <- paste0("C", colnames(se_pseudobulk))
tpm_pseudobulk <- assay(se_pseudobulk, "tpm")
tpm_pseudobulk <- tpm_pseudobulk[
!(rownames(tpm_pseudobulk) %in% mt_transcripts_isosceles),
]
tpm_pseudobulk <- t(
t(tpm_pseudobulk) / colSums(tpm_pseudobulk) * 1e6
)Prepare Isosceles bulk RNA-Seq data:
se_bulk <- readRDS(file.path(
bulk_dir, "isosceles_se_transcript.rds"
))
tpm_bulk <- assay(se_bulk, "tpm")
tpm_bulk[, 7] <- (tpm_bulk[, 7] + tpm_bulk[, 8]) / 2
tpm_bulk <- tpm_bulk[, 1:7]
colnames(tpm_bulk) <- bulk_sample_names
tpm_bulk <- tpm_bulk[, bulk_sample_order]
tpm_bulk <- tpm_bulk[
!(rownames(tpm_bulk) %in% mt_transcripts_isosceles),
]
tpm_bulk <- t(
t(tpm_bulk) / colSums(tpm_bulk) * 1e6
)Select top transcripts for further analysis:
top_transcripts <- get_hvts(sc_counts)Re-normalize the TPM values:
tpm_bulk <- tpm_bulk[top_transcripts,]
tpm_pseudobulk <- tpm_pseudobulk[top_transcripts,]
tpm_bulk <- t(
t(tpm_bulk) / colSums(tpm_bulk) * 1e6
)
tpm_pseudobulk <- t(
t(tpm_pseudobulk) / colSums(tpm_pseudobulk) * 1e6
)Calculate the correlation matrices:
cor_spearman <- cor(tpm_pseudobulk[top_transcripts,],
tpm_bulk[top_transcripts,],
method = "spearman")
mean_rel_diff <- get_mean_rel_diff_matrix(
tpm_pseudobulk[top_transcripts,],
tpm_bulk[top_transcripts,]
)Spearman correlation matrix:
as.data.frame(cor_spearman)Spearman correlation heatmap:
pheatmap(
cor_spearman,
color = heatmap_palette(100, direction = heatmap_palette_direction),
breaks = seq(min_cor_spearman, max_cor_spearman, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
Mean relative difference matrix:
as.data.frame(mean_rel_diff)Mean relative difference heatmap:
pheatmap(
mean_rel_diff,
color = heatmap_palette(100, direction = -1),
breaks = seq(min_mean_rel_diff, max_mean_rel_diff, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
TPM scatter plots:
scatterplot_df <- prepare_scatterplot_data(tpm_bulk,
tpm_pseudobulk)
scatterplot_df$tool <- "Isosceles"
scatterplot_df_list[["Isosceles"]] <- scatterplot_df
create_scatterplot_paired(scatterplot_df)
create_scatterplot_combined(scatterplot_df)
IsoQuant bulk-pseudobulk correlation plots
Prepare IsoQuant pseudobulk RNA-Seq data:
isoquant_sc_df <- read_delim(file.path(
sc_dir, "isoquant_transcript_grouped_tpm.tsv"
))
isoquant_sc_transcript_ids <- isoquant_sc_df[, 1, drop = TRUE]
isoquant_sc_counts <- as(as.matrix(isoquant_sc_df[, c(-1)]), "dgCMatrix")
rownames(isoquant_sc_counts) <- isoquant_sc_transcript_ids
isoquant_sc_counts <- isoquant_sc_counts[
!(rownames(isoquant_sc_counts) %in% mt_transcripts_ensembl),
]
cell_labels <- souporcell_cluster[colnames(isoquant_sc_counts)]
cell_selector <- cell_labels %in% c("0", "1", "2")
isoquant_sc_counts <- isoquant_sc_counts[, cell_selector]
cell_labels <- cell_labels[cell_selector]
isoquant_pseudobulk_tpm <- t(rowsum(t(isoquant_sc_counts), cell_labels))
colnames(isoquant_pseudobulk_tpm) <- paste0(
"C", colnames(isoquant_pseudobulk_tpm)
)
isoquant_pseudobulk_tpm <- t(
t(isoquant_pseudobulk_tpm) / colSums(isoquant_pseudobulk_tpm) * 1e6
)Prepare IsoQuant bulk RNA-Seq data:
isoquant_bulk_list <- lapply(bulk_sample_ids, function(sample_id) {
isoquant_df <- read_delim(file.path(
bulk_dir, glue("isoquant_{sample_id}_transcript_tpm.tsv")
))
})
isoquant_bulk_tx_ids <- unique(unlist(sapply(
isoquant_bulk_list, function(df) {df[, 1, drop = TRUE]}
)))
isoquant_bulk_tpm <- sapply(isoquant_bulk_list, function(df) {
tpm_values <- df$TPM
names(tpm_values) <- df[, 1, drop = TRUE]
tpm_values <- tpm_values[isoquant_bulk_tx_ids]
tpm_values[is.na(tpm_values)] <- 0
return(tpm_values)
})
rownames(isoquant_bulk_tpm) <- isoquant_bulk_tx_ids
isoquant_bulk_tpm[, 7] <-
(isoquant_bulk_tpm[, 7] + isoquant_bulk_tpm[, 8]) / 2
isoquant_bulk_tpm <- isoquant_bulk_tpm[, 1:7]
colnames(isoquant_bulk_tpm) <- bulk_sample_names
isoquant_bulk_tpm <- isoquant_bulk_tpm[, bulk_sample_order]
isoquant_bulk_tpm <- fill_missing_matrix(isoquant_bulk_tpm,
rownames(isoquant_pseudobulk_tpm))
isoquant_bulk_tpm <- t(
t(isoquant_bulk_tpm) / colSums(isoquant_bulk_tpm) * 1e6
)Select top transcripts for further analysis:
top_transcripts <- get_hvts(isoquant_sc_counts)Re-normalize the TPM values:
isoquant_bulk_tpm <- isoquant_bulk_tpm[top_transcripts,]
isoquant_pseudobulk_tpm <- isoquant_pseudobulk_tpm[top_transcripts,]
isoquant_bulk_tpm <- t(
t(isoquant_bulk_tpm) / colSums(isoquant_bulk_tpm) * 1e6
)
isoquant_pseudobulk_tpm <- t(
t(isoquant_pseudobulk_tpm) / colSums(isoquant_pseudobulk_tpm) * 1e6
)Calculate the correlation matrices:
cor_spearman <- cor(isoquant_pseudobulk_tpm[top_transcripts,],
isoquant_bulk_tpm[top_transcripts,],
method = "spearman")
mean_rel_diff <- get_mean_rel_diff_matrix(
isoquant_pseudobulk_tpm[top_transcripts,],
isoquant_bulk_tpm[top_transcripts,]
)Spearman correlation matrix:
as.data.frame(cor_spearman)Spearman correlation heatmap:
pheatmap(
cor_spearman,
color = heatmap_palette(100, direction = heatmap_palette_direction),
breaks = seq(min_cor_spearman, max_cor_spearman, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
Mean relative difference matrix:
as.data.frame(mean_rel_diff)Mean relative difference heatmap:
pheatmap(
mean_rel_diff,
color = heatmap_palette(100, direction = -1),
breaks = seq(min_mean_rel_diff, max_mean_rel_diff, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
TPM scatter plots:
scatterplot_df <- prepare_scatterplot_data(isoquant_bulk_tpm,
isoquant_pseudobulk_tpm)
scatterplot_df$tool <- "IsoQuant"
scatterplot_df_list[["IsoQuant"]] <- scatterplot_df
create_scatterplot_paired(scatterplot_df)
create_scatterplot_combined(scatterplot_df)
FLAMES bulk-pseudobulk correlation plots
Prepare FLAMES pseudobulk RNA-Seq data:
flames_sc_df <- read_csv(file.path(
sc_dir, "flames_transcript_count.csv.gz"
))
flames_sc_df <- flames_sc_df[!grepl("_", flames_sc_df$transcript_id),]
flames_sc_transcript_ids <- flames_sc_df$transcript_id
flames_sc_counts <- as(as.matrix(flames_sc_df[, c(-1, -2)]), "dgCMatrix")
rownames(flames_sc_counts) <- flames_sc_transcript_ids
flames_sc_counts <- flames_sc_counts[
!(rownames(flames_sc_counts) %in% mt_transcripts_ensembl),
]
cell_labels <- souporcell_cluster[colnames(flames_sc_counts)]
cell_selector <- cell_labels %in% c("0", "1", "2")
flames_sc_counts <- flames_sc_counts[, cell_selector]
cell_labels <- cell_labels[cell_selector]
flames_pseudobulk_tpm <- t(rowsum(t(flames_sc_counts), cell_labels))
colnames(flames_pseudobulk_tpm) <- paste0(
"C", colnames(flames_pseudobulk_tpm)
)
flames_pseudobulk_tpm <- t(
t(flames_pseudobulk_tpm) / colSums(flames_pseudobulk_tpm) * 1e6
)Prepare FLAMES bulk RNA-Seq data:
flames_bulk_list <- lapply(bulk_sample_ids, function(sample_id) {
flames_df <- read_csv(file.path(
bulk_dir, glue("flames_{sample_id}_transcript_count.csv.gz")
))
flames_df <- flames_df[!grepl("_", flames_df$transcript_id),]
return(flames_df)
})
flames_bulk_tx_ids <- unique(unlist(sapply(
flames_bulk_list, function(df) {df$transcript_id}
)))
flames_bulk_tpm <- sapply(flames_bulk_list, function(df) {
tpm_values <- df[, 3, drop = TRUE]
names(tpm_values) <- df$transcript_id
tpm_values <- tpm_values[flames_bulk_tx_ids]
tpm_values[is.na(tpm_values)] <- 0
return(tpm_values)
})
rownames(flames_bulk_tpm) <- flames_bulk_tx_ids
flames_bulk_tpm <- t(t(flames_bulk_tpm) / colSums(flames_bulk_tpm) * 1e6)
flames_bulk_tpm[, 7] <- (flames_bulk_tpm[, 7] + flames_bulk_tpm[, 8]) / 2
flames_bulk_tpm <- flames_bulk_tpm[, 1:7]
colnames(flames_bulk_tpm) <- bulk_sample_names
flames_bulk_tpm <- flames_bulk_tpm[, bulk_sample_order]
flames_bulk_tpm <- fill_missing_matrix(flames_bulk_tpm,
rownames(flames_pseudobulk_tpm))
flames_bulk_tpm <- t(
t(flames_bulk_tpm) / colSums(flames_bulk_tpm) * 1e6
)Select top transcripts for further analysis:
top_transcripts <- get_hvts(flames_sc_counts)Re-normalize the TPM values:
flames_bulk_tpm <- flames_bulk_tpm[top_transcripts,]
flames_pseudobulk_tpm <- flames_pseudobulk_tpm[top_transcripts,]
flames_bulk_tpm <- t(
t(flames_bulk_tpm) / colSums(flames_bulk_tpm) * 1e6
)
flames_pseudobulk_tpm <- t(
t(flames_pseudobulk_tpm) / colSums(flames_pseudobulk_tpm) * 1e6
)Calculate the correlation matrices:
cor_spearman <- cor(flames_pseudobulk_tpm[top_transcripts,],
flames_bulk_tpm[top_transcripts,],
method = "spearman")
mean_rel_diff <- get_mean_rel_diff_matrix(
flames_pseudobulk_tpm[top_transcripts,],
flames_bulk_tpm[top_transcripts,]
)Spearman correlation matrix:
as.data.frame(cor_spearman)Spearman correlation heatmap:
pheatmap(
cor_spearman,
color = heatmap_palette(100, direction = heatmap_palette_direction),
breaks = seq(min_cor_spearman, max_cor_spearman, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
Mean relative difference matrix:
as.data.frame(mean_rel_diff)Mean relative difference heatmap:
pheatmap(
mean_rel_diff,
color = heatmap_palette(100, direction = -1),
breaks = seq(min_mean_rel_diff, max_mean_rel_diff, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
TPM scatter plots:
scatterplot_df <- prepare_scatterplot_data(flames_bulk_tpm,
flames_pseudobulk_tpm)
scatterplot_df$tool <- "FLAMES"
scatterplot_df_list[["FLAMES"]] <- scatterplot_df
create_scatterplot_paired(scatterplot_df)
create_scatterplot_combined(scatterplot_df)
Sicelore bulk-pseudobulk correlation plots
Prepare Sicelore pseudobulk RNA-Seq data:
sicelore_sc_df <- read_delim(file.path(
sc_dir, "sicelore_isomatrix.txt"
))
sicelore_sc_df <- sicelore_sc_df[sicelore_sc_df$transcriptId != "undef",]
sicelore_sc_transcript_ids <- sicelore_sc_df$transcriptId
sicelore_sc_counts <- as(as.matrix(sicelore_sc_df[, c(-1, -2, -3)]),
"dgCMatrix")
rownames(sicelore_sc_counts) <- sicelore_sc_transcript_ids
sicelore_sc_counts <- sicelore_sc_counts[
!(rownames(sicelore_sc_counts) %in% mt_transcripts_ensembl),
]
cell_labels <- souporcell_cluster[colnames(sicelore_sc_counts)]
cell_selector <- cell_labels %in% c("0", "1", "2")
sicelore_sc_counts <- sicelore_sc_counts[, cell_selector]
cell_labels <- cell_labels[cell_selector]
sicelore_pseudobulk_tpm <- t(rowsum(t(sicelore_sc_counts), cell_labels))
colnames(sicelore_pseudobulk_tpm) <- paste0(
"C", colnames(sicelore_pseudobulk_tpm)
)
sicelore_pseudobulk_tpm <- t(
t(sicelore_pseudobulk_tpm) / colSums(sicelore_pseudobulk_tpm) * 1e6
)Prepare Sicelore bulk RNA-Seq data:
sicelore_bulk_list <- lapply(bulk_sample_ids, function(sample_id) {
sicelore_df <- read_delim(file.path(
bulk_dir, glue("sicelore_{sample_id}_sicelore_isomatrix.txt")
))
sicelore_df <- sicelore_df[sicelore_df$transcriptId != "undef",]
return(sicelore_df)
})
sicelore_bulk_tx_ids <- unique(unlist(sapply(
sicelore_bulk_list, function(df) {df$transcriptId}
)))
sicelore_bulk_tpm <- sapply(sicelore_bulk_list, function(df) {
tpm_values <- df[, 4, drop = TRUE]
names(tpm_values) <- df$transcriptId
tpm_values <- tpm_values[sicelore_bulk_tx_ids]
tpm_values[is.na(tpm_values)] <- 0
return(tpm_values)
})
rownames(sicelore_bulk_tpm) <- sicelore_bulk_tx_ids
sicelore_bulk_tpm <- t(
t(sicelore_bulk_tpm) / colSums(sicelore_bulk_tpm) * 1e6
)
sicelore_bulk_tpm[, 7] <-
(sicelore_bulk_tpm[, 7] + sicelore_bulk_tpm[, 8]) / 2
sicelore_bulk_tpm <- sicelore_bulk_tpm[, 1:7]
colnames(sicelore_bulk_tpm) <- bulk_sample_names
sicelore_bulk_tpm <- sicelore_bulk_tpm[, bulk_sample_order]
sicelore_bulk_tpm <- fill_missing_matrix(sicelore_bulk_tpm,
rownames(sicelore_pseudobulk_tpm))
sicelore_bulk_tpm <- t(
t(sicelore_bulk_tpm) / colSums(sicelore_bulk_tpm) * 1e6
)Select top transcripts for further analysis:
top_transcripts <- get_hvts(sicelore_sc_counts)Re-normalize the TPM values:
sicelore_bulk_tpm <- sicelore_bulk_tpm[top_transcripts,]
sicelore_pseudobulk_tpm <- sicelore_pseudobulk_tpm[top_transcripts,]
sicelore_bulk_tpm <- t(
t(sicelore_bulk_tpm) / colSums(sicelore_bulk_tpm) * 1e6
)
sicelore_pseudobulk_tpm <- t(
t(sicelore_pseudobulk_tpm) / colSums(sicelore_pseudobulk_tpm) * 1e6
)Calculate the correlation matrices:
cor_spearman <- cor(sicelore_pseudobulk_tpm[top_transcripts,],
sicelore_bulk_tpm[top_transcripts,],
method = "spearman")
mean_rel_diff <- get_mean_rel_diff_matrix(
sicelore_pseudobulk_tpm[top_transcripts,],
sicelore_bulk_tpm[top_transcripts,]
)Spearman correlation matrix:
as.data.frame(cor_spearman)Spearman correlation heatmap:
pheatmap(
cor_spearman,
color = heatmap_palette(100, direction = heatmap_palette_direction),
breaks = seq(min_cor_spearman, max_cor_spearman, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
Mean relative difference matrix:
as.data.frame(mean_rel_diff)Mean relative difference heatmap:
pheatmap(
mean_rel_diff,
color = heatmap_palette(100, direction = -1),
breaks = seq(min_mean_rel_diff, max_mean_rel_diff, length.out = 101),
cluster_rows = FALSE, cluster_cols = FALSE
)
TPM scatter plots:
scatterplot_df <- prepare_scatterplot_data(sicelore_bulk_tpm,
sicelore_pseudobulk_tpm)
scatterplot_df$tool <- "Sicelore"
scatterplot_df_list[["Sicelore"]] <- scatterplot_df
create_scatterplot_paired(scatterplot_df)
create_scatterplot_combined(scatterplot_df)
Summary barplots
Prepare summary barplot data:
plot_df <- do.call(rbind, scatterplot_df_list)
rownames(plot_df) <- NULL
plot_df$tool <- factor(plot_df$tool, levels = c("Isosceles", "IsoQuant",
"FLAMES", "Sicelore"))
plot_df <- plot_df %>%
group_by(tool, sample_bulk, sample_pseudobulk) %>%
summarise(
status = unique(status),
corr_spearman = cor(tpm_bulk, tpm_pseudobulk,
method = "spearman"),
mean_rel_diff = get_mean_rel_diff(tpm_bulk, tpm_pseudobulk)
) %>%
ungroup()
plot_df <- plot_df %>%
group_by(tool, status) %>%
summarise(
corr_spearman_avg = mean(corr_spearman),
corr_spearman_se = sd(corr_spearman) / sqrt(n()),
mean_rel_diff_avg = mean(mean_rel_diff),
mean_rel_diff_se = sd(mean_rel_diff) / sqrt(n())
) %>%
ungroup()Summary barplot (Spearman correlation):
ggplot(plot_df, mapping = aes(x = tool,
y = corr_spearman_avg,
fill = status)) +
geom_col(position = position_dodge(), color = "black", width = 0.5) +
geom_errorbar(
mapping = aes(ymin = corr_spearman_avg - corr_spearman_se,
ymax = corr_spearman_avg + corr_spearman_se),
width = 0.1,
color = "black",
position = position_dodge(0.5)
) +
scale_fill_manual(values = c(`Correct vs Correct` = "grey",
`Correct vs Decoy` = "darkred")) +
coord_cartesian(ylim = c(0, 1)) +
labs(y = "Correlation",
fill = "") +
theme_classic() +
theme(legend.position = "top",
legend.key.size = unit(0.4, "cm"),
aspect.ratio = 1,
axis.title.x = element_blank(),
axis.text.x = element_text(size = 8))
Summary barplot (mean relative difference):
ggplot(plot_df, mapping = aes(x = tool,
y = mean_rel_diff_avg,
fill = status)) +
geom_col(position = position_dodge(), color = "black", width = 0.5) +
geom_errorbar(
mapping = aes(ymin = mean_rel_diff_avg - mean_rel_diff_se,
ymax = mean_rel_diff_avg + mean_rel_diff_se),
width = 0.1,
color = "black",
position = position_dodge(0.5)
) +
scale_fill_manual(values = c(`Correct vs Correct` = "grey",
`Correct vs Decoy` = "darkred")) +
coord_cartesian(ylim = c(0, max_mean_rel_diff + 0.1)) +
labs(y = "Mean rel. diff.",
fill = "") +
theme_classic() +
theme(legend.position = "top",
legend.key.size = unit(0.4, "cm"),
aspect.ratio = 1,
axis.title.x = element_blank(),
axis.text.x = element_text(size = 8))
Metric difference and CI table:
get_diff_value <- function(x) {
return(abs(x[1] - x[2]))
}
get_diff_sd <- function(x) {
return(sqrt(sum(x ** 2)))
}
get_ci_factor <- function(x) {
qnorm(1 - (1 - x)/2)
}
diff_df <- plot_df %>%
group_by(tool) %>%
summarise(
corr_spearman = get_diff_value(corr_spearman_avg),
corr_spearman_upper_ci_95 = get_diff_value(corr_spearman_avg) +
(get_diff_sd(corr_spearman_se) * get_ci_factor(0.95)),
corr_spearman_lower_ci_95 = get_diff_value(corr_spearman_avg) -
(get_diff_sd(corr_spearman_se) * get_ci_factor(0.95)),
corr_spearman_upper_ci_90 = get_diff_value(corr_spearman_avg) +
(get_diff_sd(corr_spearman_se) * get_ci_factor(0.90)),
corr_spearman_lower_ci_90 = get_diff_value(corr_spearman_avg) -
(get_diff_sd(corr_spearman_se) * get_ci_factor(0.90)),
corr_spearman_upper_ci_68 = get_diff_value(corr_spearman_avg) +
(get_diff_sd(corr_spearman_se) * get_ci_factor(0.68)),
corr_spearman_lower_ci_68 = get_diff_value(corr_spearman_avg) -
(get_diff_sd(corr_spearman_se) * get_ci_factor(0.68)),
mean_rel_diff = get_diff_value(mean_rel_diff_avg),
mean_rel_diff_upper_ci_95 = get_diff_value(mean_rel_diff_avg) +
(get_diff_sd(mean_rel_diff_se) * get_ci_factor(0.95)),
mean_rel_diff_lower_ci_95 = get_diff_value(mean_rel_diff_avg) -
(get_diff_sd(mean_rel_diff_se) * get_ci_factor(0.95)),
mean_rel_diff_upper_ci_90 = get_diff_value(mean_rel_diff_avg) +
(get_diff_sd(mean_rel_diff_se) * get_ci_factor(0.90)),
mean_rel_diff_lower_ci_90 = get_diff_value(mean_rel_diff_avg) -
(get_diff_sd(mean_rel_diff_se) * get_ci_factor(0.90)),
mean_rel_diff_upper_ci_68 = get_diff_value(mean_rel_diff_avg) +
(get_diff_sd(mean_rel_diff_se) * get_ci_factor(0.68)),
mean_rel_diff_lower_ci_68 = get_diff_value(mean_rel_diff_avg) -
(get_diff_sd(mean_rel_diff_se) * get_ci_factor(0.68))
)
diff_df <- as.data.frame(diff_df)
rownames(diff_df) <- diff_df$tool
diff_df$tool <- NULL
diff_df <- as.data.frame(t(diff_df))
write.csv(diff_df, file.path(csv_dir, glue("data_{top_n_hvts}_hvts.csv")))
diff_dfSession Info
sessionInfo()## R version 4.2.1 (2022-06-23)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.4 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/liblapack.so.3
##
## locale:
## [1] C
##
## attached base packages:
## [1] stats4 stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] viridis_0.6.2 viridisLite_0.4.1
## [3] pheatmap_1.0.12 scran_1.24.1
## [5] scater_1.24.0 scuttle_1.6.3
## [7] SingleCellExperiment_1.18.1 SummarizedExperiment_1.26.1
## [9] Biobase_2.56.0 GenomicRanges_1.48.0
## [11] GenomeInfoDb_1.32.4 IRanges_2.30.1
## [13] S4Vectors_0.34.0 BiocGenerics_0.42.0
## [15] MatrixGenerics_1.8.1 matrixStats_0.62.0
## [17] Matrix_1.5-1 scales_1.2.1
## [19] glue_1.6.2 forcats_0.5.2
## [21] stringr_1.4.1 dplyr_1.0.10
## [23] purrr_0.3.5 readr_2.1.3
## [25] tidyr_1.2.1 tibble_3.1.8
## [27] ggplot2_3.3.6 tidyverse_1.3.2
##
## loaded via a namespace (and not attached):
## [1] googledrive_2.0.0 ggbeeswarm_0.6.0
## [3] colorspace_2.0-3 ellipsis_0.3.2
## [5] bluster_1.6.0 XVector_0.36.0
## [7] BiocNeighbors_1.14.0 fs_1.5.2
## [9] farver_2.1.1 bit64_4.0.5
## [11] ggrepel_0.9.1 fansi_1.0.3
## [13] lubridate_1.8.0 xml2_1.3.3
## [15] codetools_0.2-18 sparseMatrixStats_1.8.0
## [17] cachem_1.0.6 knitr_1.40
## [19] jsonlite_1.8.3 broom_1.0.1
## [21] cluster_2.1.4 dbplyr_2.2.1
## [23] compiler_4.2.1 httr_1.4.4
## [25] dqrng_0.3.0 backports_1.4.1
## [27] assertthat_0.2.1 fastmap_1.1.0
## [29] gargle_1.2.1 limma_3.52.4
## [31] cli_3.4.1 BiocSingular_1.12.0
## [33] htmltools_0.5.3 tools_4.2.1
## [35] rsvd_1.0.5 igraph_1.3.5
## [37] gtable_0.3.1 GenomeInfoDbData_1.2.8
## [39] Rcpp_1.0.9 cellranger_1.1.0
## [41] jquerylib_0.1.4 vctrs_0.5.0
## [43] DelayedMatrixStats_1.18.2 xfun_0.34
## [45] beachmat_2.12.0 rvest_1.0.3
## [47] lifecycle_1.0.3 irlba_2.3.5.1
## [49] statmod_1.4.37 googlesheets4_1.0.1
## [51] edgeR_3.38.4 zlibbioc_1.42.0
## [53] vroom_1.6.0 hms_1.1.2
## [55] parallel_4.2.1 RColorBrewer_1.1-3
## [57] yaml_2.3.6 gridExtra_2.3
## [59] sass_0.4.2 stringi_1.7.8
## [61] highr_0.9 ScaledMatrix_1.4.1
## [63] BiocParallel_1.30.4 rlang_1.0.6
## [65] pkgconfig_2.0.3 bitops_1.0-7
## [67] evaluate_0.17 lattice_0.20-45
## [69] labeling_0.4.2 bit_4.0.4
## [71] tidyselect_1.2.0 magrittr_2.0.3
## [73] R6_2.5.1 generics_0.1.3
## [75] metapod_1.4.0 DelayedArray_0.22.0
## [77] DBI_1.1.3 pillar_1.8.1
## [79] haven_2.5.1 withr_2.5.0
## [81] RCurl_1.98-1.9 modelr_0.1.9
## [83] crayon_1.5.2 utf8_1.2.2
## [85] tzdb_0.3.0 rmarkdown_2.17
## [87] locfit_1.5-9.6 grid_4.2.1
## [89] readxl_1.4.1 reprex_2.0.2
## [91] digest_0.6.30 munsell_0.5.0
## [93] beeswarm_0.4.0 vipor_0.4.5
## [95] bslib_0.4.0